Semantic Knowledge Management System (SKMS)

In her book, Ontology Pipeline: A framework for building knowledge infrastructures, author Jessica Talisman describes the notion of a "semantic knowledge management system" (SKMS).

Paraphrasing, a semantic knowledge management system is an organized structure of concepts, definitions, and relationships that lets both people and machines interpret information consistently and act on that information with confidence.

A semantic knowledge management system (SKMS) is essentially the institutional memory of an enterprise; formalized, organized, defragmented, structured, machine‑interpretable, and governed so both humans and AI can act on it with confidence. At a high level, it is the shift from “knowledge stored in people’s heads and scattered documents” to "knowledge engineered as a durable, computable asset".

In the past, humans would "dip" into the well of knowledge.  Now, because of artificial intelligence, both humans and machines will be "dipping" into that well of knowledge. To maximize the potential upside of what artificial intelligence; that well of knowledge needs to be refactored.

This recast semantic knowledge will reconfigure the economics of human memory. This semantic knowledge will change business models and products.  Artificial intelligence is going to change every enterprise, large and small. Accountants will change from being "data janitors" and "spreadsheet monkeys" to being the curators and stewards of that semantic knowledge, I believe. The "spine" or "skeleton" of every enterprise knowledge graph will be driven by its very reliable and well controlled compliance reporting system.

The future of accounting, auditing, and analysis is not artificial intelligence taking over. That simply will not happen.  The future will be the skills, knowledge, and judgement of accountants, auditors, and analysts being augmented, supplemented, enhanced, and amplified by artificial intelligence.

When you hire a new employee, do you trust that employee right off the bat? No. Why would you trust an artificial intelligence application?  Trust has to be earned.  "Confidence" is earned over time. Flip the artificial intelligence equation around and all this becomes clear. What skills, knowledge, and judgement does artificial intelligence bring to the table?  Just as you learn the capabilities of a new employee over time; so to you will learn the capabilities of artificial intelligence over time.

The key to using artificial intelligence effectively is not the artificial intelligence.  The key is the skills, knowledge, and judgement that humans curate and steward; the stuff in that semantic knowledge management system.  As Talisman puts it in her book, rules help minimize the "probabilistic fog" of LLMs.

Properly configured and fed with the proper skills, knowledge, and judgment; artificial intelligence can be a very helpful tool. Artificial intelligence, like any other software, does not work by "magic". To think that or to not understand how artificial intelligence works is a recipe for disaster. If you don't understand system boundaries and if you cannot control the system; you don't have a system.

Talisman summarizes, in her book, the pieces of what she calls this semantic knowledge management system (SKMS).  For the most part, I buy into Talisman's approach.  I have also been influenced by Nicolas Figay's Inhabiting Babel, Roy Roebuck's General Endeavor Management, global open industry standards, Dave McComb's perspectives on The Future of Accounting, and several other inputs. My personal approach is AI-first, digital-first, graph-first. After extensive analysis I have synthesized everything into the following components of a semantic knowledge management system (SKMS) as applied to systems  where (a) there is no tolerance for error, (b) explain ability and traceability/trackability are important, and (c) business professionals are in charge of and operate these systems, not IT people (e.g. everything has to be easy to use). 

This is what I see grounded in Talisman's notion that we need to learn what librarians have already learned about knowledge management. The key driving use case is to minimize epistemic risk, keep risk as low as possible. The second driver is reliability and safety; there is no tolerance for error or catastrophic system failure. Zero.  Note that what you see below is an INVENTORY that needs further synthesis to make it simpler and remove duplication.
  • Controlled Vocabularies: As I have pointed out before, computers are dumb beasts. You have to give them a chance to succeed.  Part of that is agreeing on meaning and turning that meaning into understanding. This needs to be done as a collective, usually across several or many different connectives inside and outside your enterprise. A common controlled vocabulary in both human readable and machine interpretable form is an absolutely necessary starting point. Governance is critical. Provenance is critical. Traceability is critical. Explainability is critical.
  • Metadata Standards: You do not live in an island. Not your team, not your department, not your  location, not your enterprise; none are islands.  Metadata for structuring, describing, and administering your semantic knowledge is critically important.  Intuitively I believe that upper level or top level ontologies are important but I also understand the current risks as these get worked out.
  • Taxonomy: A taxonomy takes a controlled vocabulary and turns it into a hierarchical structure or set of hierarchical structures.  This is about "is-a" or "general-special" or "type-subtype" and other such connections between controlled vocabulary concepts.
  • Thesaurus: A thesaurus builds on the foundation of a taxonomy and reduces ambiguity and makes the controlled vocabulary easier to use effectively.
  • Ontology: An ontology further expands on controlled vocabularies, taxonomies, thesaurus; but also introduces new capabilities. One definition of an ontology is, "an ontology is a domain model that results from a process of ontological analysis and is a semantic contract that tells the world what your worldview is about that domain and what is assumed to exist and with which nature/definition, and what are the state of affairs deemed acceptable in that domain". To simplify understanding "ontology" and to overcome limitations, I personally tend to prefer the notion of a theory.
  • Theory: A theory provides a deliberate, purposeful, carefully vetted, explicit, formal unambiguous explanation of what things makes up a system, connections between those things, conditions those things and relations must satisfy, and the goals and objectives desired by the stakeholders of the system the theory is describing. (Theory = terms + organization + schemas + business rules + process rules + inference rules + instances)
  • Model: A model is an intentional abstraction of reality: a simplified, structured way of describing a system so people can understand it, communicate about it, and solve problems together. It is not the real-world thing itself, but a useful approximation that captures the essential elements, relationships, and patterns of a subject domain. A model is an abstraction of reality according to a certain conceptualization.  Leveraging the Atomic Design Methodology can make things easier for business professionals to use. Note that fundamentally, the information that enterprises work  with tends to be multidimensional in nature. XBRL provides a multidimensional semantic information model.  The W3C Data Cube Vocabulary is a coordinate system for data.
  • Structured Information: This system is about discrete stable units of information, not data.  Think information Legos.  I call them digital information organisms.  If you do not understand the difference, then big mistakes will be made. This is about managing professional knowledge some of  which exists within an enterprise but other knowledge exists outside the enterprise such as compliance reporting rules and standards. SHACL seems excellent for constraining structures.
  • Structured Information Report Writer: Information, in the form of structures, can be "reported" using discrete stable units of information. See this example. Think model-based semantic oriented "report writer", open global industry standard,  that provides a graph of information that can be projected into human readable form using a global standard multidimensional model, but can also be reliably consumed by artificial intelligence.
  • Knowledge Graph: The general notion of a knowledge graph is a good idea, independent of any specific technical implementation. But the W3C semantic web is the gold standard and everything you do must be reconcilable to that. This graph might be RDF, LPG, PROLOG, XBRL or maybe a combination; see The Quantum Field Database.
  • Intelligence Engine: Working on all that knowledge is a neuro-semantic oriented intelligence engine. DATALOG at it's core for safety and reliability. Note that a lot of information that accountants, auditors, and analysts work with is numeric with rich relationships.  XBRL Formula provides capabilities for processing this information above and beyond what the W3C currently provides; however the W3C is working to eliminate this gap with SPARQL 1.2 RL.
  • Adaptive Technology Stack: Enterprises build a modular, adaptable, technology stack that supports multiple technology approaches, this is a "layer cake" which supports digital distributed ledgers, crypto currency, etc. Things are modular but fit together.
  • Operations Redesign: Building on a foundation that is unstable or fundamentally wrong is a recipe for disaster.  Retrofitting artificial intelligence on top of legacy workflows that are fragmented is simply not going to work. Accounting & Audit by Design (A&AD) Framework explains my thinking well. Fundamentally, the objective is to maximize the potential of artificial intelligence. Systems should be fixed by fixing processes, not remediating errors or dealing with the consequences of system failures. Systems should fail loudly. Lean Six Sigma, Agile, and other such process control techniques should be leveraged.
  • Human-AI Teaming: Roles, teams and organizational structures are designed to enable continuous collaboration between people and AI systems, defining what human contribution looks like at the frontier.  See work system; see neuro-semantic artificial intelligence with human teaming.
  • New Value Creation: Think knowledge as a product.  Think along the lines of defensible knowledge and experience moats.  Every AI-First, Digital-First, Graph-First enterprise, regardless of sector, must decide how to position intelligence integration in the market. From AI as a product feature to AI as invisible infrastructure, positioning choice determines what customers pay for, where value accrues and how the business competes. But value will also be created within the enterprise. For example; accountants, auditors, and analysts will move from being "janitors" fixing information to the role of curators and stewards of the semantic knowledge management system.
  • Industrial Process: Building this system is hard enough without inducing accidental complexity into this system. It is intentional that a kludge should be avoided. It is deliberate that this system be elegant. It is deliberate that this system be an industrial strength process.
  • Lifecycle Management: Knowledge has a life cycle.  That lifecycle needs to be consciously managed. Knowledge is created, reviewed, approved, versioned. Conflicts need resolving. Semantic drift must be monitored. Enterprise governance is important but so is cross‑enterprise governance. Properly managed, a virtual cycle (a.k.a. feedback loop) is created.
  • Integrated: Knowledge is integrated, not fragmented: ERP, CRM, HRIS, operations, logistics, open banking, analytics. Interfaces expose semantic knowledge. Legacy systems are incrementally and systematically migrated.
  • Accountability: Accountability is built in.by design. You don't have to assemble accountability; accountability is innate.
  • Aligned with Enterprise Assurance Practices: Knowledge management is aligned with enterprise assurance practices. Metrics are identified and managed to assure semantic quality. Controls are in place to mange semantic integrity. Auditability is provided for as part of the system. Benchmarking is used to monitor AI‑assisted reasoning accuracy.
  • Human Factors and Enterprise Change Management: Effective refactoring and evolution will not occur overnight. Training and skill development needs to take place. Role evolution. Cultural adoption barriers need to be overcome. New incentive structures need to be developed.
  • Reference Implementation: Business professionals need to see and understand exactly what a semantic knowledge management system is, want it does, and the benefits that it provides. End‑to‑end examples need to be created. Sample architectures documented. Case studies created. Minimally viable design patterns need to be created.
A semantic knowledge management system is the formal, computable backbone of enterprise knowledge, enabling trustworthy AI, redesigned operations, and new forms of value creation.

Artificial intelligence is a long row to hoe. Be very, very careful of the snake oil salesmen and consultants trying to separate you from your hard earned money. Plan the work; then work the plan.

Obviously, this needs further synthesis.  What I am seeing is a hybrid. Systems that are "general" like a database that can be used for anything is not what business professionals need.  What business professionals need is a "specialized" system that sits on top of a "general" semantic knowledge management system.  Accountants, auditors, and analysts need specialized work systems build on top of more general tools; not general tools.  More to come.

Even better than a Semantic Knowledge Management System (SKMS) is a Pragmatic Knowledge Management System (PKMS); see the pragmatic web.

Additional Information:

Comments

Post a Comment

Popular posts from this blog

Reference Reporting Frameworks

Image
A financial reporting framework is a purposeful arrangement of financial line items, totals, subtotals, aggregations, disaggregations, structures, metadata, categorization, classification, assertions, restrictions, constraints, other conditions, etc. A financial reporting framework represented in a digital format should not be semantically impoverished. Leaving information out, consciously or unconsciously, is not desirable. These reference reporting frameworks provide the minimum of what is necessary. They are modest yet not impoverished. A financial reporting framework provides the elements necessary for reporting organized by basic line item element, superordinate elements, and subordinate elements organized by type per the process of classification .  Financial reporting frameworks tend to start with the high level concepts of the basic accounting equation: assets, liabilities, equity; and then build out from that basic accounting equation. The following are a set of reference ...
Read more

Accounting & Audit by Design (A&AD) Framework

Image
The Accounting & Audit by Design (A&AD) Framework is about freedom from drudgery for accountants, auditors, and analysts. It is about streamlining processes. A&AD is about eliminating the waste of time and energy that goes into transaction chasing . It is about eliminating the need to resort to things like " the plug ".  A&AD is about fixing processes that allow mistakes rather than spending time fixing mistakes a process allowed or dealing with the consequences of a mistake escaping from the system undetected. The following sections will help you understand what the Accounting & Audit by Design (A&AD) Framework is and what it enables. Environment What is A&AD? Why A&AD Now? How does A&AD Work? Business Outcome Fifty to seventy five year ago it would have been impossible to contemplate something like A&AD; it just was not feasible. We did the best we could to implement accounting information systems using computers and software that we...
Read more

Overview

Image
( Executive summary ) Innovation doesn’t arrive. Innovation evolves. Imagine it.  Freedom from drudgery. Freedom from the toil of tedious, repetitive, mundane, monotonous tasks. Far less time needs to be spent fulfilling the role of "data janitor".  Time spent "transaction chasing" is significantly reduced. Accounting is a well understood system which has evolved over many thousands of years. Accounting, reporting, auditing, and analysis is being digitized. This digitization will enable processes to be industrialized  and productivity boosts to be realized. If you are an accountant, auditor, or analyst it is time to update your tradecraft. This overview is an attempt to take the work that I have done to understand XBRL-based reports and back into a framework for explaining those reports, accounting and audit working papers, analysis models to accountants and software engineers building software for accountants.  I am now referring to this as model-driven reporting ...
Read more